Jin Daily AI Trivia – NVIDIA shipped their first CPU-only System - Vera
NVIDIA’s Ian Buck literally hand-delivered the first-ever #Vera CPU systems to OpenAI, Anthropic, SpaceXAI, and Oracle Cloud.
Important point: this is #not the full Vera CPU + Rubin GPU combo yet.
This is NVIDIA shipping the #CPU-only side first.
Before Vera, NVIDIA already had Grace CPU. Grace was built for AI servers, HPC, and Grace Hopper systems, with strong efficiency and tight GPU integration. But it never really felt like NVIDIA was trying to challenge Intel Xeon or AMD EPYC directly as a data center CPU platform (focus on power efficiency)
Vera is different.
Vera is an 88-core CPU using NVIDIA’s own Olympus cores, with up to #1.2TB/s memory bandwidth and up to #1.5TB LPDDR5X memory.
For comparison:
- AMD EPYC 9005 / Turin: around 614 GB/s
- Intel Xeon 6 12 Channel: around 845 GB/s
- NVIDIA Vera: up to 1.2 TB/s
That bandwidth number is the real story.
Modern AI data centers are no longer just limited by GPU compute. Once the GPUs become insanely fast, the CPU side needs to keep up with scheduling, data movement, compilation, scripting, analytics, agent sandboxing, and all the boring but critical work around inference.
This is where Vera is trying to stand out.
NVIDIA’s first CPU that can seriously push beyond traditional x86 server platforms in memory bandwidth.
NVIDIA is no longer just selling GPUs into servers.
They are slowly turning the whole AI factory into an NVIDIA platform - CPU, GPU, NVLink, networking, rack design, and software stack.
Hope you learn something new today. See ya!
